Integrating artificial limbs as part of one's body involves complex neuroplastic changes resulting from various sensory inputs. While somatosensory feedback is crucial, plastic processes

that enable embodiment remain unknown. We investigated this using somatosensory evoked fields (SEFs) in the primary somatosensory cortex (S1) following the Rubber Hand Illusion (RHI), known

to quickly induce artificial limb embodiment. During electrical stimulation of the little finger and thumb, 19 adults underwent neuromagnetic recordings before and after the RHI. We found

early SEF displacement, including an illusion-brain correlation between extent of embodiment and specific changes to the first cortical response at 20 ms in Area 3b, within S1. Furthermore,

subsequent displacement into Area 1 may relate to early visual-tactile integration that initiates embodiment. Here we provide evidence for multiple neuroplastic processes in S1—lasting

beyond the illusion—supporting integration of artificial limbs like prostheses within the body representation.

The sense that an artificial limb has become a part of one’s body (embodiment) involves neuroplastic changes arising from multiple and sometimes conflicting sensory inputs to the brain1,2.

However, many aspects of successfully embodying new body parts, or how this might fail remain largely unknown (as occurs in up to 44% of amputees3, although he reasons for abandoning a

prosthesis are many and varied—see4 for a detailed discussion). Given that somatosensory feedback is crucial for embodiment5,6,7,8, this study sought to determine the adaptive processes

within somatosensory responses that accompany embodiment of an artificial limb.

Since the primary somatosensory cortex (S1) is the first cortical region reached by tactile afferent inputs, it is thought to have a central role in the acceptance of a new body part into

the body schema9,10,11. Current evidence suggests that S1 is attenuated in embodiment to facilitate prioritization of visual inputs1. However, it remains unclear what are the contributions

of S1 to the accommodation of an artificial limb into the body representation.

S1 is known to undergo significant neuroplasticity in response to use or environmental changes (i.e., bottom-up processes), balanced with changes in expectations (i.e., top-down

processes)12. Neuroplastic cortical remapping following loss of function, such as losing a body part, is well-documented (e.g.13,14,15,16). One way to investigate this type of remapping

employs peripheral electrical somatosensory stimulation, that causes a relay of activity along the somatosensory pathway towards the cortex (e.g.17). Signals at specific latencies are known

to result from activity within different structures. The first cortical signal after peripheral stimulation of the wrist at about 20 ms (N20 or m20 of the somatosensory evoked potential or

field, SEP or SEF) reflects direct and indirect effects of thalamic input to Brodmann area (BA) 3b within S118. Area 3b has relatively small and well-defined receptive fields, responsible

for detailed analysis of tactile stimuli, and precise localization of the stimulus on the body map. Following this, outputs from BA3b and the thalamus arrive to BA1 as early as 25 ms (P25),

in which there are larger receptive fields that integrate sensory inputs from adjacent areas of the body, for complex perceptual judgments such as object recognition19,20,21,22.

Interestingly, it has recently been found that BA1 was more responsive to tactile stimulation when visual information was included23. While it is known that visual information is essential

for artificial limb embodiment24, the study by Rosenthal and colleagues points to the specific involvement of BA1 in visuo-tactile integration for successful embodiment.

The most-studied cortical responses to somatosensory stimulation occur at early (20–50 ms) and mid-latencies (50–100 ms). Although the effect of embodiment on displacements of the

early-latency components has not been studied, there are interesting findings for the mid-latency component which reflect the interaction between several brain regions that convey complex

stimulus information and are generally considered to represent higher cognitive processes in comparison with shorter-latency responses25,26,27. Among these mid-latency responses, an

important study found evidence for neuroplastic cortical remapping following the addition of a body part: displacement of the m60 source was observed during the illusion of having a third

arm28. However, effects of embodiment on displacements of early-latency S1 responses remain to be determined.

Early-latency cortical responses related to stimulus processing are linked to the inflow of sensory information, and they are subject to state dependant changes. For example, a displacement

of the source of m20 activity as well as the source at about 50 ms (m40) (BA3b and 1, respectively) was observed following disuse by anesthesia17. The displacement of the m40 was similarly

observed due to long term use in violin-players29. This use/disuse displacement in early SEF sources is a neuroplastic change that may be an important mechanism underlying changes to the

body representation. These early-latency responses (20–50 ms), reflecting initial stimulus processing, have been shown to contain most of the clinically relevant cortical somatosensory

response components within S130,31. For example, only components within these latencies are impacted by astereognosis32 (inability to identify objects by touch), and thus will be the focus

of the current study. Displacements at the earliest latencies could point to a functional contribution of basic fundamental somatosensory processes (e.g., stimulus encoding) within S1 to

successful embodiment.

To study the integration of an artificial limb within the body representation and its neural correlates, the Rubber Hand Illusion (RHI) is often employed in healthy adults and in

amputees33,34,35. In this paradigm, the participant experiences an illusion of owning the fake hand (i.e. embodiment) that begins within seconds. The illusion occurs when the participant

observes an artificial rubber hand stroked with a paintbrush by an experimenter, who synchronously strokes the subject’s hidden real hand36,37,38. This illusion is thought to arise from the

complex integration of bottom-up multisensory information and top-down expectations about sensory information39.

The effects of the RHI on S1 and related somatosensory areas have been studied by measuring SEPs using electroencephalography (EEG), yielding mixed results. Several studies have found that

the RHI enhanced a long-latency component at 140 ms40,41 (N140), likely originating in the secondary somatosensory cortex42. One study demonstrated that the RHI reduced a mid-latency

component around 50 ms11 (P45). Lastly, only one study demonstrated effects within the early-latency components: the RHI reduced activity within the 20–25 ms time window43 (N20-P25

component, occurring in BA3b and BA1, respectively). Although this study demonstrates a relationship between the RHI and the earliest SEPs, it remains unknown whether this attenuation of

activity is accompanied by any adaptive processes to accommodate the artificial limb.

Early SEPs or SEFs provide accurate information on the location of sensory stimuli on the body, and a relative shift in SEF source location could be an adaptive process enabling successful

embodiment. That is, a change in SEF source location following the RHI could underlie changes to the body schema, as might occur after loss or gain of a body part. Furthermore, it is known

that this illusion relies upon top-down processes to prioritize visual somatosensory information24. Therefore, the question driving the current study is whether there is a relationship

between embodiment and effects on somatosensory representation areas within S1. We expect to observe stronger changes in BA1, occuring after the earliest components at 20 ms within BA3b,

thus representing embodiment effects at the onset of integrative processes within sensory cortices, that coordinate a reduction in representation of neighbouring body parts.

Most previous studies investigating neuroplasticity associated with the embodiment of an artificial limb have relied upon EEG that has insufficient spatial resolution to observe shifts in

SEP sources, or functional magnetic resonance imaging (fMRI) that has insufficient temporal resolution. Thus, the current study was conducted using magnetoencephalography (MEG), having

millisecond-temporal and millimetre-spatial resolution44,45 to determine the relationship between the RHI and early-latency SEF source locations and magnitudes within S1. Similar to previous

studies measuring neuroplastic shifts in S1 (e.g. 17,29), subjects underwent neuromagnetic recordings during electrical stimulation of the little finger and thumb immediately before and

after the RHI, to quantify relative changes to the main early SEF components and sources due to artificial limb embodiment. The extent of embodiment was measured using validated

questionnaires38, and correlated with neuromagnetic findings.

Nineteen healthy adults participated in this experiment. All volunteers signed a written informed consent before their participation in this study. The study was approved by the local Ethics

Committees (Province of Venice and Campus Bio-Medico University of Rome) and the protocols are in accordance with the latest Declaration of Helsinki (2013)46.

The experimental procedure lasted about 30-min, and was performed as follows. All subjects completed the Edinburgh Handedness Inventory47 to assess their hand dominance. Subjects sat upright

in a comfortable armchair inside a magnetically shielded room with their eyes closed. For SEF assessments, ring electrodes were placed on the little finger and thumb (Little and Thumb,

respectively) of the left hand for each subject and session during neuromagnetic recording. Using a high voltage stimulator (DS7A, Digitimer Ltd., UK), 400 stimulus repetitions (200 per

finger) were delivered with the following parameters: square pulse duration 0.2 ms, interstimulus interval ranging from 250 to 270 ms and amplitude 3-times the sensory threshold. Stimulation

to the little finger and thumb was randomly interleaved, with the stimulator housed outside of the magnetically shielded room.

This was conducted before (Pre) and immediately after (Post) a five-minute synchronous RHI procedure (Fig. 1A). For the RHI, subjects were instructed to place their left hand on a wooden

platform, with the upper arm and shoulder covered by a towel. While the left hand was occluded from view by a vertical panel, a lifelike rubber hand was positioned next to the panel in a

visible position, at a distance of 15 cm from the subject’s left hand. Both the real (left) and rubber hands were fitted with nitrile examination gloves (Fig. 1B), to make the hands look

similar. The experimenter instructed the participants to fixate on the artificial hand for the entire duration of the RHI procedure. During the 5-min procedure, the experimenter stroked the

participant’s left and the rubber hand with two identical paintbrushes at a pace of approximately 1 Hz. Importantly, just as the somatosensory stimulation, the RHI was performed within the

magnetically shielded room, without removing the subject from the MEG. This was to ensure the comparability of pre- and post-RHI measurements.

Experimental procedure and Setup. (A) Diagram of experimental procedure. Neuromagnetic evoked activity of thumb and little fingers electrical stimulation was collected before (Pre) and after

(Post) the administration of synchronous RHI that lasted 5 min. The extent of embodiment was quantified through the RHI-index calculated from the answers to a questionnaire administered

immediately following the procedure. (B) Rubber hand illusion setup, occurring between Pre and Post SEF procedures.

To quantify the extent of self-attribution to the rubber hand, participants were provided with a nine-item questionnaire38 (Supplementary Material) with which participants were asked to rate

the extent to which the nine items did or did not apply, using a 7-point scale. For this scale, -3 meant “absolutely certain that it did not apply,” 0 meant “uncertain whether it applied or

not,” and + 3 meant “absolutely certain that it applied.” In order to control for participant suggestibility, three items in the questionnaire measure the illusion, whereas the other six

items served as control for compliance, suggestibility, and “placebo effect”. From this, the RHI-index is calculated as the difference between the mean score of the illusion items compared

with the mean score of the control items48,49, and serves as a quantification of the extent of embodiment experienced for each subject and condition.

In order to localize MEG activity to each individual’s anatomy, T1-weighted structural MR images were collected for each subject using a 3 T Ingenia CX Philips scanner (Philips Medical

Systems, Best, The Netherlands).

Neuromagnetic activity was recorded using a whole-head 275-Channel CTF MEG system (VSM MedTech Systems Inc., Coquitlam, BC, Canada) in a magnetically shielded room. Data were collected at a

rate of 1200 samples/s. Small coils placed at fiducial locations (nasion and preauricular points) were used with continuous head localization to monitor head position during recording. Head

shapes and fiducial locations were digitized using a 3D Fastrack Digitizer (Polhemus, Colchester, Vermont, USA), which was used to co-register source images to the subject’s MRI using the

Brainstorm Matlab toolbox50.

Continuously recorded MEG data were segmented into 200 epochs of 1500 ms duration (500 ms pre-stimulus baseline), each for Little and Thumb, and for each session (Pre and Post). Epochs in

which the peak-to-peak amplitude across MEG channels exceeded 3 pT during the time points of interest (− 50 to 240 ms) were automatically labeled and rejected following visual inspection,

resulting in a mean of 170 epochs (SD = 27.7) for each finger and session included for analysis. Due to excessive noise for one subject, data from 120 channels were excluded, and analysis

was performed on the remaining 153 channels for that subject.

Data were highpass filtered off-line at 0.01 Hz, with a band-stop filter (50, 100, 150 Hz) to remove power line noise. A non-phase shifting (bi-directional) 4th order Butterworth filter was

used. Mean head position was calculated offline, and a multi-sphere head model51 was used, implemented in the BrainWave Matlab toolbox52. In order to measure any differences in source

location of SEFs, an event-related beamformer53,54 with 2 mm spatial resolution was used to generate source activity images on a 3D grid (591,872 voxels, or 24 × 18 × 16 cm) for averaged

brain responses, also using BrainWave. This is a spatial filtering method that computes volumetric images of instantaneous source power corresponding to selected time points in the average

(evoked) brain responses.

Data from Pre and Post sessions were combined for each finger to compute the data covariance used in estimating the beamformer spatial filter weights from the single trial data. In order to

exclude effects from the stimulus artifact, the covariance window used was 10–240 ms from stimulus onset54. Pseudo-z statistical beamformer images were created every 1 ms over the period

between 15 and 50 ms following the stimulus. Individual global maxima within the beamformer image across space and time were determined within this time window, and source direction was

aligned across subjects, in native source space. Specifically we sought to identify canonical SEF responses at 20, 27, 35, and 45 ms within S155,56,57. Given that, for example the m20

component may occur at slightly different latencies and source coordinates between subjects, a component is the label given to the peaks of activation across conditions for a between-subject

average source and latency (e.g., the m20 component). For group averaging, spatial normalization was based on the MNI (T1) template brain, and subsequent scaling to Talairach coordinates

were carried out using SPM12 (Wellcome Centre for Human Neuroimaging, London, United Kingdom). MNI coordinates of group-averaged source locations were plotted onto the ICBM152 template brain

using Brainstorm.

For all acquired MEG and behavioural data, normality was tested using the Shapiro–Wilk test. If the data were not normally distributed, non-parametric statistical tests were used.

In order to confirm the induction of the illusion, we calculated the RHI-index as the difference between the mean scores of the illusion items and the mean scores of the control items. This

index was used as the illusion outcome for the RHI condition.

To verify that the results of the RHI questionnaire was not due to participant suggestibility, the mean score of the three items employed to measure the illusion was compared with the mean

score of the six items that served to control for compliance, suggestibility, and placebo effect using a paired Wilcoxon Signed Rank test.

For each subject and session, source locations at different latencies of Little and Thumb were used to calculate their Euclidian distance in native space.

In order to determine the within-subject effects of RHI on the Euclidian distance between SEF source locations we conducted a 2-way repeated measures ANOVA with two factors (session (2

levels: Pre vs Post) and component (3 levels: m20, m35, m45)). For each component, planned Pre vs Post comparisons were conducted using paired t-tests comparing values.

The source activity for each component evoked by the electrical stimulation was quantified. Since SEF peak magnitude values were not normally distributed, they were normalized using a

cube-root transformation. In order to determine the within-subject effect of RHI on the SEF magnitude we ran a 3-way repeated measures ANOVA with three factors [session (2 levels: Pre vs

Post), finger (2 levels: Little vs Thumb) and components (3 levels: m20, m35, m45)). For each component, planned Pre vs Post comparisons were conducted using paired t-tests comparing values.

In order to determine the presence of a relationship between embodiment measures and changes (Post- Pre) to evoked neuromagnetic activity, we conducted between-subject correlations: for each

component (m20, m27, m35, m45) we correlated the change in Euclidian distance and the magnitude with RHI-index values using Spearman’s rank correlation. Although we did not hypothesize any

specific relational differences between fingers, and therefore averaged magnitudes for Little and Thumb, post-hoc comparisons considered them separately. All statistical tests were conducted

and plots were prepared using R Statistical Software58. Corrections for multiple post-hoc comparisons were performed using Holm-adjusted values.

All 19 subjects complied with task instructions and completed the experiment. One subject (male) was excluded from analyses due to abnormalities discovered in the structural MR image. Data

from the remaining 18 (13 females, range 22–56 years) subjects were analyzed.

For the RHI (conducted between Pre and Post SEF procedures), mean RHI-index values were 4.19 ± 0.38 (Fig. 2), which are comparable to values obtained in previous studies (e.g.59), confirming

induction of the illusion (positive RHI-index values) for 17 out of 18 participants. The mean value of the illusion items was significantly higher (2.4 ± 0.15) than the mean value of the

control items (− 1.75 ± 0.31; p